Feature selection based on a crow search algorithm for big data classification

The common issues of big data are that many of the features may not be relevant. Feature selection has been proven to be an effective way to improve the results of many classification algorithms. Binary crow search algorithm (BCSA) has been developed by getting inspired from natural phenomena to perform feature selection. In BCSA, the flight length parameter plays an important role in the performance of this algorithm. To improve the classification performance with reasonably selected features, an improvement of determining the flight length parameter by employing the concept of opposition-based learning strategy of BCSA is proposed. Experimental results on two datasets show the proposed algorithm, OBL-BCSA, has an advantage over the traditional BCSA in terms of selecting relevant features with a high classification performance. Further, the performance of the OBLBCSA is compared with other algorithms in term of the computational time efficiency which is revealing that OBLBCSA outperforms them.